Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright IBM Corp. 2006
4 * Author(s): Heiko Carstens <heiko.carstens@de.ibm.com>
5 */
6
7#include <linux/memblock.h>
8#include <linux/pfn.h>
9#include <linux/mm.h>
10#include <linux/init.h>
11#include <linux/list.h>
12#include <linux/hugetlb.h>
13#include <linux/slab.h>
14#include <asm/cacheflush.h>
15#include <asm/pgalloc.h>
16#include <asm/setup.h>
17#include <asm/tlbflush.h>
18#include <asm/sections.h>
19#include <asm/set_memory.h>
20
21static DEFINE_MUTEX(vmem_mutex);
22
23static void __ref *vmem_alloc_pages(unsigned int order)
24{
25 unsigned long size = PAGE_SIZE << order;
26
27 if (slab_is_available())
28 return (void *)__get_free_pages(GFP_KERNEL, order);
29 return (void *) memblock_phys_alloc(size, size);
30}
31
32static void vmem_free_pages(unsigned long addr, int order)
33{
34 /* We don't expect boot memory to be removed ever. */
35 if (!slab_is_available() ||
36 WARN_ON_ONCE(PageReserved(phys_to_page(addr))))
37 return;
38 free_pages(addr, order);
39}
40
41void *vmem_crst_alloc(unsigned long val)
42{
43 unsigned long *table;
44
45 table = vmem_alloc_pages(CRST_ALLOC_ORDER);
46 if (table)
47 crst_table_init(table, val);
48 return table;
49}
50
51pte_t __ref *vmem_pte_alloc(void)
52{
53 unsigned long size = PTRS_PER_PTE * sizeof(pte_t);
54 pte_t *pte;
55
56 if (slab_is_available())
57 pte = (pte_t *) page_table_alloc(&init_mm);
58 else
59 pte = (pte_t *) memblock_phys_alloc(size, size);
60 if (!pte)
61 return NULL;
62 memset64((u64 *)pte, _PAGE_INVALID, PTRS_PER_PTE);
63 return pte;
64}
65
66static void vmem_pte_free(unsigned long *table)
67{
68 /* We don't expect boot memory to be removed ever. */
69 if (!slab_is_available() ||
70 WARN_ON_ONCE(PageReserved(virt_to_page(table))))
71 return;
72 page_table_free(&init_mm, table);
73}
74
75#define PAGE_UNUSED 0xFD
76
77/*
78 * The unused vmemmap range, which was not yet memset(PAGE_UNUSED) ranges
79 * from unused_pmd_start to next PMD_SIZE boundary.
80 */
81static unsigned long unused_pmd_start;
82
83static void vmemmap_flush_unused_pmd(void)
84{
85 if (!unused_pmd_start)
86 return;
87 memset(__va(unused_pmd_start), PAGE_UNUSED,
88 ALIGN(unused_pmd_start, PMD_SIZE) - unused_pmd_start);
89 unused_pmd_start = 0;
90}
91
92static void __vmemmap_use_sub_pmd(unsigned long start, unsigned long end)
93{
94 /*
95 * As we expect to add in the same granularity as we remove, it's
96 * sufficient to mark only some piece used to block the memmap page from
97 * getting removed (just in case the memmap never gets initialized,
98 * e.g., because the memory block never gets onlined).
99 */
100 memset(__va(start), 0, sizeof(struct page));
101}
102
103static void vmemmap_use_sub_pmd(unsigned long start, unsigned long end)
104{
105 /*
106 * We only optimize if the new used range directly follows the
107 * previously unused range (esp., when populating consecutive sections).
108 */
109 if (unused_pmd_start == start) {
110 unused_pmd_start = end;
111 if (likely(IS_ALIGNED(unused_pmd_start, PMD_SIZE)))
112 unused_pmd_start = 0;
113 return;
114 }
115 vmemmap_flush_unused_pmd();
116 __vmemmap_use_sub_pmd(start, end);
117}
118
119static void vmemmap_use_new_sub_pmd(unsigned long start, unsigned long end)
120{
121 void *page = __va(ALIGN_DOWN(start, PMD_SIZE));
122
123 vmemmap_flush_unused_pmd();
124
125 /* Could be our memmap page is filled with PAGE_UNUSED already ... */
126 __vmemmap_use_sub_pmd(start, end);
127
128 /* Mark the unused parts of the new memmap page PAGE_UNUSED. */
129 if (!IS_ALIGNED(start, PMD_SIZE))
130 memset(page, PAGE_UNUSED, start - __pa(page));
131 /*
132 * We want to avoid memset(PAGE_UNUSED) when populating the vmemmap of
133 * consecutive sections. Remember for the last added PMD the last
134 * unused range in the populated PMD.
135 */
136 if (!IS_ALIGNED(end, PMD_SIZE))
137 unused_pmd_start = end;
138}
139
140/* Returns true if the PMD is completely unused and can be freed. */
141static bool vmemmap_unuse_sub_pmd(unsigned long start, unsigned long end)
142{
143 void *page = __va(ALIGN_DOWN(start, PMD_SIZE));
144
145 vmemmap_flush_unused_pmd();
146 memset(__va(start), PAGE_UNUSED, end - start);
147 return !memchr_inv(page, PAGE_UNUSED, PMD_SIZE);
148}
149
150/* __ref: we'll only call vmemmap_alloc_block() via vmemmap_populate() */
151static int __ref modify_pte_table(pmd_t *pmd, unsigned long addr,
152 unsigned long end, bool add, bool direct)
153{
154 unsigned long prot, pages = 0;
155 int ret = -ENOMEM;
156 pte_t *pte;
157
158 prot = pgprot_val(PAGE_KERNEL);
159 if (!MACHINE_HAS_NX)
160 prot &= ~_PAGE_NOEXEC;
161
162 pte = pte_offset_kernel(pmd, addr);
163 for (; addr < end; addr += PAGE_SIZE, pte++) {
164 if (!add) {
165 if (pte_none(*pte))
166 continue;
167 if (!direct)
168 vmem_free_pages(pfn_to_phys(pte_pfn(*pte)), 0);
169 pte_clear(&init_mm, addr, pte);
170 } else if (pte_none(*pte)) {
171 if (!direct) {
172 void *new_page = vmemmap_alloc_block(PAGE_SIZE, NUMA_NO_NODE);
173
174 if (!new_page)
175 goto out;
176 pte_val(*pte) = __pa(new_page) | prot;
177 } else {
178 pte_val(*pte) = addr | prot;
179 }
180 } else {
181 continue;
182 }
183 pages++;
184 }
185 ret = 0;
186out:
187 if (direct)
188 update_page_count(PG_DIRECT_MAP_4K, add ? pages : -pages);
189 return ret;
190}
191
192static void try_free_pte_table(pmd_t *pmd, unsigned long start)
193{
194 pte_t *pte;
195 int i;
196
197 /* We can safely assume this is fully in 1:1 mapping & vmemmap area */
198 pte = pte_offset_kernel(pmd, start);
199 for (i = 0; i < PTRS_PER_PTE; i++, pte++) {
200 if (!pte_none(*pte))
201 return;
202 }
203 vmem_pte_free(__va(pmd_deref(*pmd)));
204 pmd_clear(pmd);
205}
206
207/* __ref: we'll only call vmemmap_alloc_block() via vmemmap_populate() */
208static int __ref modify_pmd_table(pud_t *pud, unsigned long addr,
209 unsigned long end, bool add, bool direct)
210{
211 unsigned long next, prot, pages = 0;
212 int ret = -ENOMEM;
213 pmd_t *pmd;
214 pte_t *pte;
215
216 prot = pgprot_val(SEGMENT_KERNEL);
217 if (!MACHINE_HAS_NX)
218 prot &= ~_SEGMENT_ENTRY_NOEXEC;
219
220 pmd = pmd_offset(pud, addr);
221 for (; addr < end; addr = next, pmd++) {
222 next = pmd_addr_end(addr, end);
223 if (!add) {
224 if (pmd_none(*pmd))
225 continue;
226 if (pmd_large(*pmd) && !add) {
227 if (IS_ALIGNED(addr, PMD_SIZE) &&
228 IS_ALIGNED(next, PMD_SIZE)) {
229 if (!direct)
230 vmem_free_pages(pmd_deref(*pmd), get_order(PMD_SIZE));
231 pmd_clear(pmd);
232 pages++;
233 } else if (!direct && vmemmap_unuse_sub_pmd(addr, next)) {
234 vmem_free_pages(pmd_deref(*pmd), get_order(PMD_SIZE));
235 pmd_clear(pmd);
236 }
237 continue;
238 }
239 } else if (pmd_none(*pmd)) {
240 if (IS_ALIGNED(addr, PMD_SIZE) &&
241 IS_ALIGNED(next, PMD_SIZE) &&
242 MACHINE_HAS_EDAT1 && addr && direct &&
243 !debug_pagealloc_enabled()) {
244 pmd_val(*pmd) = addr | prot;
245 pages++;
246 continue;
247 } else if (!direct && MACHINE_HAS_EDAT1) {
248 void *new_page;
249
250 /*
251 * Use 1MB frames for vmemmap if available. We
252 * always use large frames even if they are only
253 * partially used. Otherwise we would have also
254 * page tables since vmemmap_populate gets
255 * called for each section separately.
256 */
257 new_page = vmemmap_alloc_block(PMD_SIZE, NUMA_NO_NODE);
258 if (new_page) {
259 pmd_val(*pmd) = __pa(new_page) | prot;
260 if (!IS_ALIGNED(addr, PMD_SIZE) ||
261 !IS_ALIGNED(next, PMD_SIZE)) {
262 vmemmap_use_new_sub_pmd(addr, next);
263 }
264 continue;
265 }
266 }
267 pte = vmem_pte_alloc();
268 if (!pte)
269 goto out;
270 pmd_populate(&init_mm, pmd, pte);
271 } else if (pmd_large(*pmd)) {
272 if (!direct)
273 vmemmap_use_sub_pmd(addr, next);
274 continue;
275 }
276 ret = modify_pte_table(pmd, addr, next, add, direct);
277 if (ret)
278 goto out;
279 if (!add)
280 try_free_pte_table(pmd, addr & PMD_MASK);
281 }
282 ret = 0;
283out:
284 if (direct)
285 update_page_count(PG_DIRECT_MAP_1M, add ? pages : -pages);
286 return ret;
287}
288
289static void try_free_pmd_table(pud_t *pud, unsigned long start)
290{
291 const unsigned long end = start + PUD_SIZE;
292 pmd_t *pmd;
293 int i;
294
295 /* Don't mess with any tables not fully in 1:1 mapping & vmemmap area */
296 if (end > VMALLOC_START)
297 return;
298#ifdef CONFIG_KASAN
299 if (start < KASAN_SHADOW_END && KASAN_SHADOW_START > end)
300 return;
301#endif
302 pmd = pmd_offset(pud, start);
303 for (i = 0; i < PTRS_PER_PMD; i++, pmd++)
304 if (!pmd_none(*pmd))
305 return;
306 vmem_free_pages(pud_deref(*pud), CRST_ALLOC_ORDER);
307 pud_clear(pud);
308}
309
310static int modify_pud_table(p4d_t *p4d, unsigned long addr, unsigned long end,
311 bool add, bool direct)
312{
313 unsigned long next, prot, pages = 0;
314 int ret = -ENOMEM;
315 pud_t *pud;
316 pmd_t *pmd;
317
318 prot = pgprot_val(REGION3_KERNEL);
319 if (!MACHINE_HAS_NX)
320 prot &= ~_REGION_ENTRY_NOEXEC;
321 pud = pud_offset(p4d, addr);
322 for (; addr < end; addr = next, pud++) {
323 next = pud_addr_end(addr, end);
324 if (!add) {
325 if (pud_none(*pud))
326 continue;
327 if (pud_large(*pud)) {
328 if (IS_ALIGNED(addr, PUD_SIZE) &&
329 IS_ALIGNED(next, PUD_SIZE)) {
330 pud_clear(pud);
331 pages++;
332 }
333 continue;
334 }
335 } else if (pud_none(*pud)) {
336 if (IS_ALIGNED(addr, PUD_SIZE) &&
337 IS_ALIGNED(next, PUD_SIZE) &&
338 MACHINE_HAS_EDAT2 && addr && direct &&
339 !debug_pagealloc_enabled()) {
340 pud_val(*pud) = addr | prot;
341 pages++;
342 continue;
343 }
344 pmd = vmem_crst_alloc(_SEGMENT_ENTRY_EMPTY);
345 if (!pmd)
346 goto out;
347 pud_populate(&init_mm, pud, pmd);
348 } else if (pud_large(*pud)) {
349 continue;
350 }
351 ret = modify_pmd_table(pud, addr, next, add, direct);
352 if (ret)
353 goto out;
354 if (!add)
355 try_free_pmd_table(pud, addr & PUD_MASK);
356 }
357 ret = 0;
358out:
359 if (direct)
360 update_page_count(PG_DIRECT_MAP_2G, add ? pages : -pages);
361 return ret;
362}
363
364static void try_free_pud_table(p4d_t *p4d, unsigned long start)
365{
366 const unsigned long end = start + P4D_SIZE;
367 pud_t *pud;
368 int i;
369
370 /* Don't mess with any tables not fully in 1:1 mapping & vmemmap area */
371 if (end > VMALLOC_START)
372 return;
373#ifdef CONFIG_KASAN
374 if (start < KASAN_SHADOW_END && KASAN_SHADOW_START > end)
375 return;
376#endif
377
378 pud = pud_offset(p4d, start);
379 for (i = 0; i < PTRS_PER_PUD; i++, pud++) {
380 if (!pud_none(*pud))
381 return;
382 }
383 vmem_free_pages(p4d_deref(*p4d), CRST_ALLOC_ORDER);
384 p4d_clear(p4d);
385}
386
387static int modify_p4d_table(pgd_t *pgd, unsigned long addr, unsigned long end,
388 bool add, bool direct)
389{
390 unsigned long next;
391 int ret = -ENOMEM;
392 p4d_t *p4d;
393 pud_t *pud;
394
395 p4d = p4d_offset(pgd, addr);
396 for (; addr < end; addr = next, p4d++) {
397 next = p4d_addr_end(addr, end);
398 if (!add) {
399 if (p4d_none(*p4d))
400 continue;
401 } else if (p4d_none(*p4d)) {
402 pud = vmem_crst_alloc(_REGION3_ENTRY_EMPTY);
403 if (!pud)
404 goto out;
405 p4d_populate(&init_mm, p4d, pud);
406 }
407 ret = modify_pud_table(p4d, addr, next, add, direct);
408 if (ret)
409 goto out;
410 if (!add)
411 try_free_pud_table(p4d, addr & P4D_MASK);
412 }
413 ret = 0;
414out:
415 return ret;
416}
417
418static void try_free_p4d_table(pgd_t *pgd, unsigned long start)
419{
420 const unsigned long end = start + PGDIR_SIZE;
421 p4d_t *p4d;
422 int i;
423
424 /* Don't mess with any tables not fully in 1:1 mapping & vmemmap area */
425 if (end > VMALLOC_START)
426 return;
427#ifdef CONFIG_KASAN
428 if (start < KASAN_SHADOW_END && KASAN_SHADOW_START > end)
429 return;
430#endif
431
432 p4d = p4d_offset(pgd, start);
433 for (i = 0; i < PTRS_PER_P4D; i++, p4d++) {
434 if (!p4d_none(*p4d))
435 return;
436 }
437 vmem_free_pages(pgd_deref(*pgd), CRST_ALLOC_ORDER);
438 pgd_clear(pgd);
439}
440
441static int modify_pagetable(unsigned long start, unsigned long end, bool add,
442 bool direct)
443{
444 unsigned long addr, next;
445 int ret = -ENOMEM;
446 pgd_t *pgd;
447 p4d_t *p4d;
448
449 if (WARN_ON_ONCE(!PAGE_ALIGNED(start | end)))
450 return -EINVAL;
451 for (addr = start; addr < end; addr = next) {
452 next = pgd_addr_end(addr, end);
453 pgd = pgd_offset_k(addr);
454
455 if (!add) {
456 if (pgd_none(*pgd))
457 continue;
458 } else if (pgd_none(*pgd)) {
459 p4d = vmem_crst_alloc(_REGION2_ENTRY_EMPTY);
460 if (!p4d)
461 goto out;
462 pgd_populate(&init_mm, pgd, p4d);
463 }
464 ret = modify_p4d_table(pgd, addr, next, add, direct);
465 if (ret)
466 goto out;
467 if (!add)
468 try_free_p4d_table(pgd, addr & PGDIR_MASK);
469 }
470 ret = 0;
471out:
472 if (!add)
473 flush_tlb_kernel_range(start, end);
474 return ret;
475}
476
477static int add_pagetable(unsigned long start, unsigned long end, bool direct)
478{
479 return modify_pagetable(start, end, true, direct);
480}
481
482static int remove_pagetable(unsigned long start, unsigned long end, bool direct)
483{
484 return modify_pagetable(start, end, false, direct);
485}
486
487/*
488 * Add a physical memory range to the 1:1 mapping.
489 */
490static int vmem_add_range(unsigned long start, unsigned long size)
491{
492 return add_pagetable(start, start + size, true);
493}
494
495/*
496 * Remove a physical memory range from the 1:1 mapping.
497 */
498static void vmem_remove_range(unsigned long start, unsigned long size)
499{
500 remove_pagetable(start, start + size, true);
501}
502
503/*
504 * Add a backed mem_map array to the virtual mem_map array.
505 */
506int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
507 struct vmem_altmap *altmap)
508{
509 int ret;
510
511 mutex_lock(&vmem_mutex);
512 /* We don't care about the node, just use NUMA_NO_NODE on allocations */
513 ret = add_pagetable(start, end, false);
514 if (ret)
515 remove_pagetable(start, end, false);
516 mutex_unlock(&vmem_mutex);
517 return ret;
518}
519
520void vmemmap_free(unsigned long start, unsigned long end,
521 struct vmem_altmap *altmap)
522{
523 mutex_lock(&vmem_mutex);
524 remove_pagetable(start, end, false);
525 mutex_unlock(&vmem_mutex);
526}
527
528void vmem_remove_mapping(unsigned long start, unsigned long size)
529{
530 mutex_lock(&vmem_mutex);
531 vmem_remove_range(start, size);
532 mutex_unlock(&vmem_mutex);
533}
534
535int vmem_add_mapping(unsigned long start, unsigned long size)
536{
537 int ret;
538
539 if (start + size > VMEM_MAX_PHYS ||
540 start + size < start)
541 return -ERANGE;
542
543 mutex_lock(&vmem_mutex);
544 ret = vmem_add_range(start, size);
545 if (ret)
546 vmem_remove_range(start, size);
547 mutex_unlock(&vmem_mutex);
548 return ret;
549}
550
551/*
552 * map whole physical memory to virtual memory (identity mapping)
553 * we reserve enough space in the vmalloc area for vmemmap to hotplug
554 * additional memory segments.
555 */
556void __init vmem_map_init(void)
557{
558 struct memblock_region *reg;
559
560 for_each_memblock(memory, reg)
561 vmem_add_range(reg->base, reg->size);
562 __set_memory((unsigned long)_stext,
563 (unsigned long)(_etext - _stext) >> PAGE_SHIFT,
564 SET_MEMORY_RO | SET_MEMORY_X);
565 __set_memory((unsigned long)_etext,
566 (unsigned long)(__end_rodata - _etext) >> PAGE_SHIFT,
567 SET_MEMORY_RO);
568 __set_memory((unsigned long)_sinittext,
569 (unsigned long)(_einittext - _sinittext) >> PAGE_SHIFT,
570 SET_MEMORY_RO | SET_MEMORY_X);
571 __set_memory(__stext_dma, (__etext_dma - __stext_dma) >> PAGE_SHIFT,
572 SET_MEMORY_RO | SET_MEMORY_X);
573
574 /* we need lowcore executable for our LPSWE instructions */
575 set_memory_x(0, 1);
576
577 pr_info("Write protected kernel read-only data: %luk\n",
578 (unsigned long)(__end_rodata - _stext) >> 10);
579}
1/*
2 * arch/s390/mm/vmem.c
3 *
4 * Copyright IBM Corp. 2006
5 * Author(s): Heiko Carstens <heiko.carstens@de.ibm.com>
6 */
7
8#include <linux/bootmem.h>
9#include <linux/pfn.h>
10#include <linux/mm.h>
11#include <linux/module.h>
12#include <linux/list.h>
13#include <linux/hugetlb.h>
14#include <linux/slab.h>
15#include <asm/pgalloc.h>
16#include <asm/pgtable.h>
17#include <asm/setup.h>
18#include <asm/tlbflush.h>
19#include <asm/sections.h>
20
21static DEFINE_MUTEX(vmem_mutex);
22
23struct memory_segment {
24 struct list_head list;
25 unsigned long start;
26 unsigned long size;
27};
28
29static LIST_HEAD(mem_segs);
30
31static void __ref *vmem_alloc_pages(unsigned int order)
32{
33 if (slab_is_available())
34 return (void *)__get_free_pages(GFP_KERNEL, order);
35 return alloc_bootmem_pages((1 << order) * PAGE_SIZE);
36}
37
38static inline pud_t *vmem_pud_alloc(void)
39{
40 pud_t *pud = NULL;
41
42#ifdef CONFIG_64BIT
43 pud = vmem_alloc_pages(2);
44 if (!pud)
45 return NULL;
46 clear_table((unsigned long *) pud, _REGION3_ENTRY_EMPTY, PAGE_SIZE * 4);
47#endif
48 return pud;
49}
50
51static inline pmd_t *vmem_pmd_alloc(void)
52{
53 pmd_t *pmd = NULL;
54
55#ifdef CONFIG_64BIT
56 pmd = vmem_alloc_pages(2);
57 if (!pmd)
58 return NULL;
59 clear_table((unsigned long *) pmd, _SEGMENT_ENTRY_EMPTY, PAGE_SIZE * 4);
60#endif
61 return pmd;
62}
63
64static pte_t __ref *vmem_pte_alloc(unsigned long address)
65{
66 pte_t *pte;
67
68 if (slab_is_available())
69 pte = (pte_t *) page_table_alloc(&init_mm, address);
70 else
71 pte = alloc_bootmem(PTRS_PER_PTE * sizeof(pte_t));
72 if (!pte)
73 return NULL;
74 clear_table((unsigned long *) pte, _PAGE_TYPE_EMPTY,
75 PTRS_PER_PTE * sizeof(pte_t));
76 return pte;
77}
78
79/*
80 * Add a physical memory range to the 1:1 mapping.
81 */
82static int vmem_add_mem(unsigned long start, unsigned long size, int ro)
83{
84 unsigned long address;
85 pgd_t *pg_dir;
86 pud_t *pu_dir;
87 pmd_t *pm_dir;
88 pte_t *pt_dir;
89 pte_t pte;
90 int ret = -ENOMEM;
91
92 for (address = start; address < start + size; address += PAGE_SIZE) {
93 pg_dir = pgd_offset_k(address);
94 if (pgd_none(*pg_dir)) {
95 pu_dir = vmem_pud_alloc();
96 if (!pu_dir)
97 goto out;
98 pgd_populate(&init_mm, pg_dir, pu_dir);
99 }
100
101 pu_dir = pud_offset(pg_dir, address);
102 if (pud_none(*pu_dir)) {
103 pm_dir = vmem_pmd_alloc();
104 if (!pm_dir)
105 goto out;
106 pud_populate(&init_mm, pu_dir, pm_dir);
107 }
108
109 pte = mk_pte_phys(address, __pgprot(ro ? _PAGE_RO : 0));
110 pm_dir = pmd_offset(pu_dir, address);
111
112#ifdef __s390x__
113 if (MACHINE_HAS_HPAGE && !(address & ~HPAGE_MASK) &&
114 (address + HPAGE_SIZE <= start + size) &&
115 (address >= HPAGE_SIZE)) {
116 pte_val(pte) |= _SEGMENT_ENTRY_LARGE;
117 pmd_val(*pm_dir) = pte_val(pte);
118 address += HPAGE_SIZE - PAGE_SIZE;
119 continue;
120 }
121#endif
122 if (pmd_none(*pm_dir)) {
123 pt_dir = vmem_pte_alloc(address);
124 if (!pt_dir)
125 goto out;
126 pmd_populate(&init_mm, pm_dir, pt_dir);
127 }
128
129 pt_dir = pte_offset_kernel(pm_dir, address);
130 *pt_dir = pte;
131 }
132 ret = 0;
133out:
134 flush_tlb_kernel_range(start, start + size);
135 return ret;
136}
137
138/*
139 * Remove a physical memory range from the 1:1 mapping.
140 * Currently only invalidates page table entries.
141 */
142static void vmem_remove_range(unsigned long start, unsigned long size)
143{
144 unsigned long address;
145 pgd_t *pg_dir;
146 pud_t *pu_dir;
147 pmd_t *pm_dir;
148 pte_t *pt_dir;
149 pte_t pte;
150
151 pte_val(pte) = _PAGE_TYPE_EMPTY;
152 for (address = start; address < start + size; address += PAGE_SIZE) {
153 pg_dir = pgd_offset_k(address);
154 pu_dir = pud_offset(pg_dir, address);
155 if (pud_none(*pu_dir))
156 continue;
157 pm_dir = pmd_offset(pu_dir, address);
158 if (pmd_none(*pm_dir))
159 continue;
160
161 if (pmd_huge(*pm_dir)) {
162 pmd_clear(pm_dir);
163 address += HPAGE_SIZE - PAGE_SIZE;
164 continue;
165 }
166
167 pt_dir = pte_offset_kernel(pm_dir, address);
168 *pt_dir = pte;
169 }
170 flush_tlb_kernel_range(start, start + size);
171}
172
173/*
174 * Add a backed mem_map array to the virtual mem_map array.
175 */
176int __meminit vmemmap_populate(struct page *start, unsigned long nr, int node)
177{
178 unsigned long address, start_addr, end_addr;
179 pgd_t *pg_dir;
180 pud_t *pu_dir;
181 pmd_t *pm_dir;
182 pte_t *pt_dir;
183 pte_t pte;
184 int ret = -ENOMEM;
185
186 start_addr = (unsigned long) start;
187 end_addr = (unsigned long) (start + nr);
188
189 for (address = start_addr; address < end_addr; address += PAGE_SIZE) {
190 pg_dir = pgd_offset_k(address);
191 if (pgd_none(*pg_dir)) {
192 pu_dir = vmem_pud_alloc();
193 if (!pu_dir)
194 goto out;
195 pgd_populate(&init_mm, pg_dir, pu_dir);
196 }
197
198 pu_dir = pud_offset(pg_dir, address);
199 if (pud_none(*pu_dir)) {
200 pm_dir = vmem_pmd_alloc();
201 if (!pm_dir)
202 goto out;
203 pud_populate(&init_mm, pu_dir, pm_dir);
204 }
205
206 pm_dir = pmd_offset(pu_dir, address);
207 if (pmd_none(*pm_dir)) {
208 pt_dir = vmem_pte_alloc(address);
209 if (!pt_dir)
210 goto out;
211 pmd_populate(&init_mm, pm_dir, pt_dir);
212 }
213
214 pt_dir = pte_offset_kernel(pm_dir, address);
215 if (pte_none(*pt_dir)) {
216 unsigned long new_page;
217
218 new_page =__pa(vmem_alloc_pages(0));
219 if (!new_page)
220 goto out;
221 pte = pfn_pte(new_page >> PAGE_SHIFT, PAGE_KERNEL);
222 *pt_dir = pte;
223 }
224 }
225 memset(start, 0, nr * sizeof(struct page));
226 ret = 0;
227out:
228 flush_tlb_kernel_range(start_addr, end_addr);
229 return ret;
230}
231
232/*
233 * Add memory segment to the segment list if it doesn't overlap with
234 * an already present segment.
235 */
236static int insert_memory_segment(struct memory_segment *seg)
237{
238 struct memory_segment *tmp;
239
240 if (seg->start + seg->size > VMEM_MAX_PHYS ||
241 seg->start + seg->size < seg->start)
242 return -ERANGE;
243
244 list_for_each_entry(tmp, &mem_segs, list) {
245 if (seg->start >= tmp->start + tmp->size)
246 continue;
247 if (seg->start + seg->size <= tmp->start)
248 continue;
249 return -ENOSPC;
250 }
251 list_add(&seg->list, &mem_segs);
252 return 0;
253}
254
255/*
256 * Remove memory segment from the segment list.
257 */
258static void remove_memory_segment(struct memory_segment *seg)
259{
260 list_del(&seg->list);
261}
262
263static void __remove_shared_memory(struct memory_segment *seg)
264{
265 remove_memory_segment(seg);
266 vmem_remove_range(seg->start, seg->size);
267}
268
269int vmem_remove_mapping(unsigned long start, unsigned long size)
270{
271 struct memory_segment *seg;
272 int ret;
273
274 mutex_lock(&vmem_mutex);
275
276 ret = -ENOENT;
277 list_for_each_entry(seg, &mem_segs, list) {
278 if (seg->start == start && seg->size == size)
279 break;
280 }
281
282 if (seg->start != start || seg->size != size)
283 goto out;
284
285 ret = 0;
286 __remove_shared_memory(seg);
287 kfree(seg);
288out:
289 mutex_unlock(&vmem_mutex);
290 return ret;
291}
292
293int vmem_add_mapping(unsigned long start, unsigned long size)
294{
295 struct memory_segment *seg;
296 int ret;
297
298 mutex_lock(&vmem_mutex);
299 ret = -ENOMEM;
300 seg = kzalloc(sizeof(*seg), GFP_KERNEL);
301 if (!seg)
302 goto out;
303 seg->start = start;
304 seg->size = size;
305
306 ret = insert_memory_segment(seg);
307 if (ret)
308 goto out_free;
309
310 ret = vmem_add_mem(start, size, 0);
311 if (ret)
312 goto out_remove;
313 goto out;
314
315out_remove:
316 __remove_shared_memory(seg);
317out_free:
318 kfree(seg);
319out:
320 mutex_unlock(&vmem_mutex);
321 return ret;
322}
323
324/*
325 * map whole physical memory to virtual memory (identity mapping)
326 * we reserve enough space in the vmalloc area for vmemmap to hotplug
327 * additional memory segments.
328 */
329void __init vmem_map_init(void)
330{
331 unsigned long ro_start, ro_end;
332 unsigned long start, end;
333 int i;
334
335 ro_start = ((unsigned long)&_stext) & PAGE_MASK;
336 ro_end = PFN_ALIGN((unsigned long)&_eshared);
337 for (i = 0; i < MEMORY_CHUNKS && memory_chunk[i].size > 0; i++) {
338 start = memory_chunk[i].addr;
339 end = memory_chunk[i].addr + memory_chunk[i].size;
340 if (start >= ro_end || end <= ro_start)
341 vmem_add_mem(start, end - start, 0);
342 else if (start >= ro_start && end <= ro_end)
343 vmem_add_mem(start, end - start, 1);
344 else if (start >= ro_start) {
345 vmem_add_mem(start, ro_end - start, 1);
346 vmem_add_mem(ro_end, end - ro_end, 0);
347 } else if (end < ro_end) {
348 vmem_add_mem(start, ro_start - start, 0);
349 vmem_add_mem(ro_start, end - ro_start, 1);
350 } else {
351 vmem_add_mem(start, ro_start - start, 0);
352 vmem_add_mem(ro_start, ro_end - ro_start, 1);
353 vmem_add_mem(ro_end, end - ro_end, 0);
354 }
355 }
356}
357
358/*
359 * Convert memory chunk array to a memory segment list so there is a single
360 * list that contains both r/w memory and shared memory segments.
361 */
362static int __init vmem_convert_memory_chunk(void)
363{
364 struct memory_segment *seg;
365 int i;
366
367 mutex_lock(&vmem_mutex);
368 for (i = 0; i < MEMORY_CHUNKS; i++) {
369 if (!memory_chunk[i].size)
370 continue;
371 seg = kzalloc(sizeof(*seg), GFP_KERNEL);
372 if (!seg)
373 panic("Out of memory...\n");
374 seg->start = memory_chunk[i].addr;
375 seg->size = memory_chunk[i].size;
376 insert_memory_segment(seg);
377 }
378 mutex_unlock(&vmem_mutex);
379 return 0;
380}
381
382core_initcall(vmem_convert_memory_chunk);